CONFIG_CPU_FREQ change

CONFIG_CPU_FREQ change

[Jan Beulich]

To my surprise, c/s 8888 enables CPU_FREQ for x86-64 rather than disabling it for i386. Did anyone at your end actually
test that if enabled this at least builds properly now? Not to mention that of course this also should work... If I
remember right, the main reason for posting a patch to disable it on 32-bits (similar to how it was on 64-bits before)
was that there were some missing symbols, and I don't think I saw any changesets addressing this. Also, from previous
discussion I seem to recall that it was generally agreed that there is little point in allowing a single domain (even
dom0) to decide whether/what power management actions should be taken without knowing about the requirements of the rest
of the system...

Thanks, Jan

Re: CONFIG_CPU_FREQ change

[Christian Limpach]

On 2/20/06, Jan Beulich <JBeulich@novell.com> wrote:
> To my surprise, c/s 8888 enables CPU_FREQ for x86-64 rather than disabling it for i386. Did anyone at your end actually
> test that if enabled this at least builds properly now? Not to mention that of course this also should work... If I
> remember right, the main reason for posting a patch to disable it on 32-bits (similar to how it was on 64-bits before)
> was that there were some missing symbols, and I don't think I saw any changesets addressing this. Also, from previous
> discussion I seem to recall that it was generally agreed that there is little point in allowing a single domain (even
> dom0) to decide whether/what power management actions should be taken without knowing about the requirements of the rest
> of the system...

I went with the final statement in the thread where you posted the
patch, from Jeremy Katz stating, that it was working for dom0 and that
disabling it would remove functionality.  It is disabled for
unprivileged guests.  I'm happy to disable it entirely, if it doesn't
build or if it doesn't work.

    christian

Re: CONFIG_CPU_FREQ change

[Jeremy Katz]

On Mon, 2006-02-20 at 18:12 +0000, Christian Limpach wrote:
> On 2/20/06, Jan Beulich <JBeulich@novell.com> wrote:
> > To my surprise, c/s 8888 enables CPU_FREQ for x86-64 rather than disabling it for i386. Did anyone at your end actually
> > test that if enabled this at least builds properly now? Not to mention that of course this also should work... If I
> > remember right, the main reason for posting a patch to disable it on 32-bits (similar to how it was on 64-bits before)
> > was that there were some missing symbols, and I don't think I saw any changesets addressing this. Also, from previous
> > discussion I seem to recall that it was generally agreed that there is little point in allowing a single domain (even
> > dom0) to decide whether/what power management actions should be taken without knowing about the requirements of the rest
> > of the system...
> 
> I went with the final statement in the thread where you posted the
> patch, from Jeremy Katz stating, that it was working for dom0 and that
> disabling it would remove functionality.  It is disabled for
> unprivileged guests.  I'm happy to disable it entirely, if it doesn't
> build or if it doesn't work.

centrino cpufreq was working on my laptop about a month ago (with the
caveat of there's not a governor that takes into account activity in a
domU, but that's fine for at least manual scaling

Jeremy

Re: CONFIG_CPU_FREQ change

[Jan Beulich]

>>> christian.limpach@gmail.com 20.02.06 19:12:30 >>>
>On 2/20/06, Jan Beulich <JBeulich@novell.com> wrote:
>> To my surprise, c/s 8888 enables CPU_FREQ for x86-64 rather than disabling it for i386. Did anyone at your end
actually
>> test that if enabled this at least builds properly now? Not to mention that of course this also should work... If I
>> remember right, the main reason for posting a patch to disable it on 32-bits (similar to how it was on 64-bits
before)
>> was that there were some missing symbols, and I don't think I saw any changesets addressing this. Also, from
previous
>> discussion I seem to recall that it was generally agreed that there is little point in allowing a single domain
(even
>> dom0) to decide whether/what power management actions should be taken without knowing about the requirements of the
rest
>> of the system...
>
>I went with the final statement in the thread where you posted the
>patch, from Jeremy Katz stating, that it was working for dom0 and that
>disabling it would remove functionality.  It is disabled for
>unprivileged guests.  I'm happy to disable it entirely, if it doesn't
>build or if it doesn't work.

p4-clockmod.c and speedstep-ich.c reference cpu_sibling_map, which doesn't exist in Xen kernels. There was one other
symbol missing, but I don't recall which one (nor which module it was referenced from).
As far as 'working' goes, I would assume that any respective statements refer to a dom0-only scenario only.

Jan